Abstract
Various song characteristics are used by avian listeners to assess singer quality. One honest indicator of quality could be vocal mimicry. The acoustic function hypothesis suggests that mimetic sounds could be functioning acoustically in the songs of some mimicking species. I wanted to determine whether mimicry added acoustic diversity to song and whether positioning of mimicked sounds increased acoustic contrast in the songs of European starlings (Sturnus vulgaris). I quantified eight song metrics using the songs of 19 males and compared mimetic to nonmimetic motifs. Mimicry expanded the acoustic range of song in both frequency and amplitude. However, acoustic contrast between motifs within song sequences was lower for mimetic sections than for
nonmimetic sections. Additionally, transition versatility within sequences was highest in sections without mimicry. These findings support the hypothesis that mimicry has acoustic function/s within starling song. Listeners could potentially use mimicry within song to assess singer quality. Additional study is required to determine whether mimicry has a similar function in the songs of other mimicking species. The acoustic structure of songs and the role of mimicry therein is a ripe avenue for understanding the relationship between song learning and mate choice.
Introduction
In songbirds, song is thought to function in mate attraction, mate choice, and territorial defense (e.g. Kroodsma and Byers 1991). As such, variation in song and
singing ability can allow females to compare males and demonstrate preferences. There is evidence suggesting that different song characteristics are important to the mating
preferences of different species. For example, females of some species prefer greater song output and song rate (Nowicki and Searcy 2004). In this case, increasing song output and song rate is difficult and can therefore function as an honest indicator of singer quality (Nowicki and Searcy 2004). Male quality has also been found to be correlated with other characteristics, such as song duration, timing of singing, aesthetic quality of song, and extent of mimicry, in different species (Kroodsma and Byers 1991).
Mimicry – the imitation of heterospecific and environmental sounds – could signal quality in a variety of ways. For example, male satin bowerbird (Ptilonorhynchus
violaceus) quality appears to be reflected in mimetic accuracy (Coleman et al. 2007), as it
seems to be in song sparrows (Melospiza melodia, Beecher and Brenowitz 2005). This relationship corresponds with the learning and performance hypothesis, which suggests that an individual’s ability to learn or produce mimicked sounds is important in mate
choice (Coleman et al. 2007, Dalziell et al. 2014). This hypothesis has received some support but is often overlooked in species-specific studies of mimicry, as mimetic accuracy is relatively difficult to quantify. Alternatively, mimicry could be used in a different way, such as to (1) introduce novel sounds to a repertoire, (2) expand the acoustic range of an individual’s song, or (3) create contrast in song. These potential
functions fall under the acoustic function hypothesis, which has not often been considered in studies of mimicry in birdsong.
The incorporation of sounds that are unique or novel may give singers an
produce and may therefore be favored by females (Vallet et al. 1998) and/or serve important social functions in song (Gil and Gahr 2002). For example, female canaries (Serinus canaria) prefer a song syllable with a lot of frequency modulation, which may demonstrate the function of a male’s respiratory and motor control systems (Vallet et al.
1998, Vallet and Kreutzer 1995). The quality of the Snarr note of water pipits (Anthus
spinoletta) indicates the dominance rank of males (Rehsteiner et al. 1998). Listeners
could use the difficulty of producing song or specific song components to assess singer quality (Leitao et al. 2006). Female swamp sparrows (Melospiza georgiana) prefer songs as close to the population maximum performance thresholds of frequency range and trill rate as possible (Ballentine et al. 2004). In mimicking species, mimicry could either introduce potential “sexy” syllables or allow individuals to sing more impressive songs.
Similarly, mimicry may allow singers to expand the acoustic range of their songs. For example, mimicked sounds in Northern mockingbird (Mimus polyglottos) song extend song maximum frequency and therefore frequency bandwidth (Gammon and Lyon 2017). Extending acoustic features in this way may yield a more attractive song. Hermit thrushes (Catharus guttatus) use songs of high or low frequency at different times of day (Roach et al. 2012), suggesting that song frequency parameters transmit information about singer quality. However, little is known about this potential function, and further study is needed.
Finally, the presence of sufficient contrast in song may be important to listeners. Hartshorne (1956) postulated that singers must avoid reaching a monotony threshold in song to retain listener interest, and therefore should limit repetition and lack of versatility. While this anti-monotony hypothesis initially described song rate and diversity of song
types, it may be expanded to an expectation that females prefer songs that are more heterogeneous and less repetitive, as is true for zebra finches (Taenopygia guttata, Neubauer 1999) and chaffinches (Fringilla coelebs, Leitao et al. 2006). That is, females should prefer songs that are more “interesting”. In species that sing songs composed of
strings of song units, heterogeneity may vary based on what these song units are, and how they are strung together. In mimicking species, incorporating mimetic sounds may allow individuals to increase diversity or contrast within their songs. For example, by alternating between nonmimetic and mimetic sounds, an individual could draw a listener’s attention and/or demonstrate singing prowess.
As the metabolic cost of singing is not insignificant (Oberweger and Goller 2001), songs that are more demanding in any of these three ways – through the addition of novel sounds, increased acoustic range, or increased acoustic contrast – could potentially indicate male quality. As such, mimicry could possibly influence mate choice in songbird species. I wanted to determine whether mimicry plays an acoustic role in the songs of male European starlings. Several lines of evidence support this idea. Mimicked notes are not used to increase repertoire size (Chapter 2). Motifs incorporating mimicry were repeated more often than species-specific sounds, and mimicry was significantly
associated with more stereotyped transitions, suggesting that mimicry serves a structural role in song (Chapter 3). Based on these findings, it appears that mimicry is used
differently from species-specific sounds, and is not passively incorporated into song. To determine whether mimicry serves an acoustic function in starling song, I asked two questions. First, are there quantitative differences between mimetic and nonmimetic motifs? I expected that mimetic motifs would extend the amplitude and
frequency range of starling song. Second, if there are quantitative differences, do these differences create greater contrast in sequences? I hypothesized that transitions between mimetic and nonmimetic motifs would have greater contrast in acoustic features than transitions between either two mimetic or two nonmimetic motifs. In both cases, mimicry should increase the heterogeneity of a male’s song.
Methods
European starling warbling song is composed of motifs. Each motif can be thought of as analogous to words in human speech and is composed of a set of one to ten notes repeated as a discrete unit. Each male has a repertoire of species-specific and mimetic motifs that is distinct from that of other males. Song has a clear organizational structure of three sections containing different types of motifs: several introductory
whistle motifs followed by a series of rambling, repeated, variable motifs, and concluding
with a series of high-frequency, loud terminal motifs (e.g. Adret-Hausberger and Güttinger 1984, Eens et al. 1989, Gentner and Hulse 1998, Gentner and Hulse 2000). Starlings regularly incorporate mimicry in songs, and motifs with mimicked components comprise an average of 46% of the repertoire of unique motifs of an individual male (Chapter 2). Mimicked sounds can appear in any of the three motif sections; however, mimicry is overrepresented within the variable motif section of song (Chapter 2).
I used the same recordings and motif libraries as I did for Chapters 2 and 3 for each of the 19 males.
I quantified eight acoustic traits for one exemplar of each motif, including minimum, maximum, and mean frequency, frequency range, mean and maximum
amplitude, motif duration, and number of notes or components in each motif. The number of components in each motif were counted manually. Motif duration was found with manual selection of the motif in the analysis window of the program Praat (v. 6.0.23; Boersma and Weenink 2019). I filtered out as much background noise as possible for all motifs before measuring other acoustic features. Due to the complexity of starling motifs, the noise filter could not be stringently employed. I then used the ‘show pitch’ and ‘show intensity’ analysis tools in Praat to measure frequency and amplitude features of all
motifs. To avoid measuring remaining background noise as much as possible, I often measured a motif using multiple steps. I did this by drawing analysis boxes around
specific motif components, saving all the measurements within each box, and then pooled measurements across boxes to calculate means, minima, and maxima. Frequency range was calculated as the difference between maximum and minimum frequency of a motif.
To determine the uniformity of my quantitative measurements, I measured 10 replicates of a subset of 18 motifs. I then used the R package rptR (Stoffel et al. 2017; R Core Team 2019) to calculate repeatability, or intraclass correlation, of the eight song metrics for each motif (Table 4.1). Standard error was found using 1000 bootstrap iterations and zero permutations.
Table 4.1. Repeatability of measurements of motif metrics across 10 replicates of 18 motifs +/- standard error. All repeatability scores had a p-value less than 0.0001.
mean frequency 0.591 +/- 0.095 min frequency 0.374 +/- 0.095
max frequency 0.671 +/- 0.093 frequency range 0.635 +/- 0.096 mean amplitude 0.48 +/- 0.1 max amplitude 0.492 +/- 0.1 duration 0.741 +/- 0.074 components 0.659 +/- 0.087
Once all traits were measured, I ran a linear mixed model with male as random effect with the package nlme (Pinheiro et al. 2019) in R, followed by ANOVA, to compare nonmimetic and mimetic motifs in all three song sections. This allowed me to determine whether mimetic motifs differed quantitatively from nonmimetic motifs. Final sample sizes were 224 nonmimetic whistles, 118 mimetic whistles, 365 nonmimetic and 370 mimetic variable motifs, and 175 nonmimetic, and 82 mimetic, terminal motifs.
Contrast within variable motif sequences
I only used the longest, middle section of song (the variable motifs) for contrast analysis. Both the whistle and terminal motif sections are characterized by high
frequency whistles, and mimicry is mostly added to the end of these motifs, instead of embedded within them (Chapter 2). As such, I focused on the variable motif section.
Contrast within song sequences was measured in two ways. First, I calculated differences in all eight traits between adjacent motifs, such that I had four transition types: nonmimetic to nonmimetic, nonmimetic to mimetic, mimetic to nonmimetic, and mimetic to mimetic. In my analysis, I used the absolute value of all differences in
statistical tests to focus on the size of contrast, while using the true values for
comparisons of the distributions. For contrast in max and mean frequency, max and mean amplitude, duration, and number of components, my sample sizes were: 2816
nonmimetic – nonmimetic, 1704 nonmimetic – mimetic, 1909 mimetic – nonmimetic, and 2661 mimetic – mimetic. For contrast in minimum frequency and frequency range, the sample sizes were 2810, 1699, 1907 and 2662, respectively.
Second, I calculated a transition versatility score (adapted from Gil and Slater 2000) for each song bout. Transition versatility was the number of unique transitions divided by the total number of transitions, per bout. I excluded transitions between the same motif, such as motif A → motif A, from the numerator to keep scores between 0 and 1.0. I then compared transition versatility to the number of unique nonmimetic and mimetic motifs in each bout. In total, I had transition versatility scores from 1,069 song bouts.
I determined the difference in contrast for all eight traits using a linear mixed model in R with male as random effect, followed by ANOVA and Tukey-HSD posthoc tests. I also compared distributions of contrast between the four transition types using Anderson-Darling tests in the R package kSamples (Scholz and Zhu 2019).
Finally, I determined the relationships between transition versatility score and either nonmimetic or mimetic motifs by using two linear mixed-effects models with male as random effect. I used the two models to determine the effects of number of unique, and total number of, nonmimetic and mimetic motifs in bouts.
Motif quantitative trait differences
There were significant differences in motif traits of mimetic and nonmimetic motifs in all song sections, although the specific patterns differed across song sections (Table 4.2). In the whistle section, mimetic motifs had significantly higher mean and maximum frequency, as well as frequency range (Figure 4.1). In contrast, in the variable motif section, mimetic motifs had significantly lower mean frequency and significantly lower minimum frequency. Mimicry in this section decreased the song frequency and increased motif duration. Mimetic terminal motifs, like variable motifs, had lower minimum frequency and longer duration. In this section, mimicry also increased the frequency range of motifs.
Mimetic motifs in all three sections were composed of significantly more components (notes).
Table 4.2. Quantitative trait means for mimetic and nonmimetic motif categories in the three song sections. Bolded means are significantly different from the respective other
mean in that category.
song section mimetic mean freq min freq max freq freq range mean amp max amp duration components
whistle no 3044.25 2289.40 3944.95 1655.54 65.32 74.15 0.69 1.49 yes 3517.84* 2475.60 4902.09* 2426.49* 65.93 75.32 0.78 2.32* variable no 3956.60* 2667.98 5539.17 2871.19 58.20 69.67 0.67 2.98 yes 3793.16 2391.54* 5566.91 3175.37 59.22 71.16 0.72* 3.25* terminal no 6573.43 4615.70 8057.58 3441.88 74.39 82.49 0.61 2.26 yes 6162.21 3909.03* 7994.27 4085.23* 73.68 82.96 0.74* 2.78* *All significance at p < 0.01
Figure 4.1. Mean differences in frequency range of nonmimetic and mimetic motifs of the three song sections. Frequency range was significantly higher in mimetic motifs for
whistle and terminal motifs.
Contrast within variable motif sequences
The size of the difference in mean, minimum, and maximum frequency, frequency range, and mean amplitude was significantly highest in transitions between two nonmimetic motifs (Table 4.3, Figure 4.2).
Figure 4.2. Boxplots of the difference in frequency (A) or amplitude (B) metrics for the four transition types: nonmimetic to nonmimetic, nonmimetic to mimetic, mimetic to
nonmimetic, and mimetic to mimetic.
Transitions between two nonmimetic motifs and two mimetic motifs had
significantly different levels of contrast in all quantitative traits except for duration. The two types of heterogeneous transitions (between nonmimetic and mimetic motifs) differed in contrast of maximum frequency, frequency range, mean amplitude, duration, and number of components. Mimetic to nonmimetic motif transitions had the lowest contrast scores for four of the eight motif traits.
transition type min freq mean freq max freq freq range mean amp max amp duration components mim-mim 1024.30 1029.24 1588.54 1881.11 7.51 7.48 0.23 1.05 mim-nonmim 949.79 1083.23 1499.50 1716.59 7.27 8.75 0.21 0.93 nonmim-mim 898.13 1034.95 1730.32 1971.54 8.38 7.99 0.24 1.05 nonmim-nonmim 1044.96* 1268.53* 1981.46* 1985.25* 9.41* 9.37 0.22 0.80
The contrast distributions of the four transition types were significantly different for all eight traits (e.g. duration, with the largest p-value: F-value = 4.87, p = 0.002, through max frequency: F-value = 74.5, p < 0.0001; Figure 4.3). Transition types did not, therefore, only differ in the size of the difference in traits (absolute value) between adjacent motifs, but also in the direction of the difference (raw value). Transitions
between mimetic motifs had less contrast than sequences of nonmimetic, species-specific motifs. The nonmimetic-nonmimetic difference distribution for maximum frequency (Figure 4.3A) had a trimodal shape.
Figure 4.3. Density distribution of the difference in max frequency (A) and mean amplitude (B) for the four transition types. Mimetic sequences have less contrast than
nonmimetic ones.
Contrary to our predictions, the transition versatility of song bouts was negatively related to the number of unique mimetic motifs in a bout (lme: t-value = -4.06, p < 0.01) but not to the number of unique nonmimetic motifs (lme: t-value = 1.06, p = 0.29). Thus,
Difference in max frequency (Hz)
Difference in mean amplitude (dB)
D e n si ty A B nonmim-nonmim nonmim-mim mim-nonmim mim-mim Transition type
as the number of unique mimetic motifs increased, the transition versatility decreased for a song bout (Figure 4.4). There was also less variation in transition versatility at high numbers of unique mimetic motifs (Figure 4.4B) than for nonmimetic motifs (Figure 4.4A). Total number of nonmimetic or mimetic motifs (including repetitions of the same motif) had a significant negative effect on transition versatility (lme: nonmimetic – t- value = -6.16, p = 0; mimetic – t-value = -3.81, p <0.01).
Figure 4.4. Transition versatility scores for nonmimetic (A) and mimetic (B) motifs. There is less variation in transition versatility at high numbers of unique mimetic motifs
than there is with many nonmimetic motifs. The trend in (B) is small but significant.
Discussion
Number of unique motifs in bout
Tr
an
si
ti
o
n
ve
rs
at
ili
ty
sc
o
re
A
B
The acoustic function hypothesis emphasizes ways in which singers could use mimicry to develop attractive song. Three proposed ways in which mimicry could function are by (1) introducing novel sounds to the song repertoire, (2) expanding acoustic range, and/or (3) adding contrast. Although no inferences can be made about novel sounds, mimicry did allow males to increase their acoustic range. Mimetic and nonmimetic motifs in European starling song are acoustically different. Some of the pattern is clear. Four traits (maximum amplitude, frequency range, duration, and number of components) are all extended by mimicry, in all three song sections. As mimicry often takes the form of imitated notes attached to the end of motifs (Chapter 2), it makes sense that duration and number of components increase with mimicry.
Variation in the other acoustic traits is less clear. Male starlings appear to also increase the loudness (amplitude) and vocal range (frequency range) of songs by
incorporating mimicry. Mimicry in the variable and terminal motif sections lowered the minimum frequency and increased the maximum frequency of motifs. In the whistle section, mimicry instead led to greater mean frequency of song. Different acoustic features seem to be more important in different song sections, and mimicry may emphasize these differences. Furthermore, the fact that mimicry extends the acoustic range of song may yield additional advantages.
Females of many species focus on specific aspects of a male’s singing behavior.
In several studies, females have shown preferences for higher frequency (zebra finch, Ritschard et al. 2010; rock sparrow, Petronia petronia, Nemeth et al. 2012), amplitude (dusky warbler, Phylloscopus fuscatus, Forstmeier et al. 2002), and complexity
signals of male quality. For example, stress early in life negatively affects song learning, as well as adult body size and immune function (Nowicki and Searcy 2004), which may be reflected in a less-developed song. It will take further study to determine whether mimetically-extended frequency bandwidth of song, such as was found in this study, and in Northern mockingbirds (Gammon and Lyon 2017), makes the singer more attractive to female listeners.
Our results present a puzzling pattern with respect to the role of mimicry in increasing acoustic contrast in songs. Mimetic motifs expanded the overall acoustic range of the songs; however, at the level of transitions between two consecutive motifs,
sequences that contained mimetic motifs showed reduced acoustic contrast. Furthermore, as the number of unique mimetic motifs increased, song versatility decreased in a song bout. Mimetic motifs “fit into” a song (created less contrast) than did nonmimetic,
species-specific motifs. Thus, while mimetic motifs can expand the overall spectral range of notes, starlings do not structure their song to use mimicry to emphasize contrast. Indeed, it appears that starlings are using mimicry to decrease contrast within song bouts, which indicates that contrast may be something males attempt to minimize. These
findings correspond with previous results showing that mimetic motifs are not necessarily used at key points within the song structure of starlings – e.g., at points of convergence or divergence within song sequences – although they are repeated more often than
nonmimetic motifs (Chapter 3). Female starlings prefer long song bouts (Gentner and Hulse 2000), which are more linear or stereotyped than shorter bouts (Chapter 3). As such, they may also prefer reduced contrast between neighboring motifs within bouts. Therefore, mimetic motifs affect the overall properties of a starling’s motif repertoire, but